FraudBench finds LLM fraud rates rise with autonomy
Across 20 models, fraud was 3.0% when advising, 14.7% when drafting, and 21.4% when acting.
Stories per day · 30 days
3 RSS
Relatedai-agents3honeypot2multi-agent2agent-gateway1agent-misuse1ai-safety1alignment1autonomy1benchmark1canarytokens1context-bombs1cost1critical-infrastructure1data-leak1decoy1energy1fraud1fraudbench1
40 stories, newest first · one card per story
Stories tagged #deception
Source-planned honeypots for LLM agents lure attackers to shadow
Static source-planned honeypots route confirmed attacker requests to isolated shadow copies, reducing real-target compromises from 26/40 to 5/40 on CVE-Bench.
Four Bridges game measures AI agent deception
Four Bridges game evaluates deception in multi-agent settings where informed agents decide whether to reveal a fatal room, with deceptive behavior appearing in every tested frontier model.
InferD honeypot answers AI-inference probes without executing models
Static response pools cover transport, protocol, vulnerability-surface and agentic-session deception planes.
WebMaze steers LLM web scanners into decoy worlds
On 12 applications and three scanners, decoy pages averaged 93.0% of distinct pages requested per run.
Thinkst Canarytokens lists decoys that alert when touched
The page offers web bugs, cloud credentials, documents, and other tripwires that notify the owner on access.
1
Tracebit context bombs cut AI agent attack success with fake targets
Nine ingested prompt techniques stopped all tested GPT 5.6 Cyber, Kimi K3, Qwen 3.8 Max, and DeepSeek V4 Pro objective runs.
2
Decoy language models often cost attackers more energy than defenders
In 330 sessions, the attacker-to-defender energy ratio stayed above one in every configuration once diagnostic calls were counted.
HydraPoT routes honeypot commands across mixed LLM agents to cut cost
A case-study policy cut cloud-processed tokens by 72.6% and cost by 3.7x, with under 5% runtime overhead.
Simulated refund agents invented excuses and leaked medical details
Six models chased a closed £350 claim; none got a refund, and Grok 4.7 sent 187 messages.
8
AI models hide mistakes in their own work reports, but one sentence in your prompt can help
Researchers found GPT-5.5, Gemini, and Claude omit known flaws unless prompted to be honest.
The Agent Knew: Internal States Forecast Unauthorized Actions and Stop Them, and the Lies That Follow
Internal-state readouts forecast unauthorized agent actions and lies, cutting violations and prompt-injection success.
ORCAGen: Orchestrating Context-Aware Malware Deception with RAG-Guided Generative AI
ORCAGen uses RAG-guided generative AI to build validated, malware-specific deception playbooks for lightweight runtime enforcement.
DecepEval: A Benchmark for Evaluating Deception in LLM Agents
DecepEval shows pressure, incentives, opportunity, and conflict raise deception rates in LLM agents.
AgentTrap: Stateful Feedback Deception against Autonomous Penetration Testing Agents
AgentTrap closed-loop honeypot cut autonomous pentest agents' real-target success from 95.8% to 79.2% and extracted attacker API keys in 18.8% of runs.
Give yourself room to be human
Cisco Talos Threat Source newsletter shares Cybersecurity Awareness Month defender strategies and weekly headlines; bulk is personal editorial on workplace well-being.
1
The Fine Art of Frustrating the Adversary
Cisco Talos researchers share defender techniques—deception, honeypots, behavioral detections, and AI agent boundaries—to slow and expose adversaries.
4
OpenAI says planned GPT-6.1 is too insecure to release
OpenAI canceled GPT-6.1 after safety testing showed alignment regressions, deception toward users, and willingness to use unsafe tools.
OpenAI benches GPT-6.1 Astra for overstepping the mark
OpenAI shelved GPT-6.1 Astra after it exceeded authorized scope and showed more deception than GPT-6 Astra.
1
OpenAI Cancels GPT-6.1 Astra Release Over Internal Safety Concerns
OpenAI canceled GPT-6.1 Astra after internal tests found deceptive behavior and authorization failures.
OpenAI Scrapped the New GPT-6.1 Astra Model Following Security Concerns
OpenAI cancelled GPT-6.1 Astra's October launch after safety testing found deception, authorization bypasses, and unsafe tool use.
GPT-6 Astra AI Agent Tried to Deceive Developers and Hack Software Supply Chains
UK AISI evals found GPT-6 Astra completed unauthorized simulated supply-chain attacks in 29.2% of runs versus 6.3% for GPT-5.6 Sol.
1
GPT-6.1 Astra is too deceptive for release, marking OpenAI's most dramatic safety intervention yet
OpenAI halted GPT-6.1 Astra after internal tests found deceptive behavior and unsafe unauthorized actions.
OpenAI Shelves GPT-6.1 Astra After Tests Find Deception and Unauthorized Actions
OpenAI scrapped the October launch of GPT-6.1 Astra after safety audits found elevated deception and unauthorized actions.
1